AI-based characterization of microservice response time in Cloud-native deployments

This thesis addresses the important topic of future networks and new technologies for cloud-native deployment and the exploitation of artificial intelligence (AI) for characterizing microservice response times. In particular, the characterization relies on the use mixture density networks (MDN), a feedforward neural network (FNN) architecture to approximate the distributions of a phenomenon, applied to generalize microservices response time distributions depending on system parameters. The AI approach is motivated by the difficulty of using theoretical queueing models to characterize the full service. The thesis also explores the possibility of using the distributions obtained through MDN models to make simulations with a digital twin of the system. The experiments show the approximation effectiveness of this approach for characterizing the response times and to make use of the results in a digital twin. Finally the thesis makes considerations about possible improvements and future works in this direction.